Re: [avr-gcc-list] push r1, pop r0

[David Brown]

On 09/11/17 04:57, Szikra Istvan wrote:
> Hi
> 
> Thanks for the SP, I missed that. 
> And apparently Atmel Studio also cannot find it, and underlines it with
> red error marker.
> It does compile, and I have found it in avr/common.h, it's probably a
> problem with __AVR_ARCH__ handling by AS...
> I guess that's what I get for trusting the IDE:)
> 
> I know that just reading SP is not enough, I do also use stack
> watermarking. It's just an additional diagnostic information.
> Note: SP can also be used to re-mark the unused stack to trace stack
> usage over time...
> (or mark stack on platforms without .init* sections.)
> 
> Note #999: 
> "the problem is in your code."
> This isn't actually my code. My code was written for ARM and looked
> something like this:
> unsigned int GetStackPointer() {
>     volatile register unsigned int sp asm("r13");
>     return sp;
> }
> that someone ported to AVR, and now I'm fixing it... ;)
> 

That code is wrong for the ARM too.  You should not fix a variable to an
assembly register that is used specifically by the compiler.

Depending on the ARM in question, you should probably use a CMSIS
function like _get_PSP().  The standard definition for gcc and Cortex-M is:

__attribute__( ( always_inline ) ) __STATIC_INLINE uint32_t __get_PSP(void)
{
  register uint32_t result;

  __ASM volatile ("MRS %0, psp\n"  : "=r" (result) );
  return(result);
}


Just to read the r13 register, the code would be something like this (I
haven't checked the details here) :

static inline uint32_t GetStackPointer(void)
{
        uint32_t sp;
        asm volatile("mov %0, r13" : "=r" (sp));
        return sp;
}


And you can probably just use __builtin_frame_address() - that should,
in theory, work on the AVR and the ARM (I have not tested it on either
target).